538635 - [injection] SPARC jit extremely slow running ecma3/Statements/etry_008.as

Status: VERIFIED FIXED

(Tamarin Graveyard :: Baseline JIT (CodegenLIR), defect, P2)

Description

[Brent Baker]

Running ecma3/Statements/etry_008.abc has started to timeout the build system when the JIT is enabled on solaris.

Previously the test would execute in under a second and now it is taking 20+ minutes. If you run with -Dinterp it executes normal (fast).

Working to try and isolate the exact change that caused the problem, there are about 6 changes to nanojit in a row.

Comment 1

[Brent Baker]

This has been tracked down to changeset 3670, which is for bug #538000

Comment 2

[Brent Baker]

changeset:   3475:3d878354f272

Marking the test to be skipped on sparc acceptance, except when running the interp. This needs to be reverted when this issue is resolved.

Comment 3

[Brent Baker]

An additional failure was also found in ecma3/Date/e15_9_5_4.as

This was marked as an expectedfail in changeset 3482:51ddd273c385

Comment 4

[Krzysztof Palacz]

TR won't build on Solaris/SPARC without the fix from 547232. 
With the fix applied to tip, can't reproduce the issue (both tests complete promptly in release and debug-debugger builds).

Comment 5

[Brent Baker]

Verified that testcase ecma3/Statements/etry_008 is running properly on solaris SPARC in tr-argo 3812.

I had previously ran all of the tests with tr-argo 3800 + nanojit patch for #547232 and the ecma3/Date/e15_9_5_4.as testcase passed. However now in tr-argo it is crashing. This is likely due to the updated nanojit that is now in tr-argo. I am verifying this bug as fixed and adding the date failure to bug #551173